Skip to main content
NCBI home page
Cardiovascular Digital Health Journal logoLink to Cardiovascular Digital Health Journal
. 2023 Jan 27;4(2):41–47. doi: 10.1016/j.cvdhj.2023.01.003

Assessment of the atrial fibrillation burden in Holter electrocardiogram recordings using artificial intelligence

Elisa Hennings ∗,†, Michael Coslovsky ‡, Rebecca E Paladini ∗,†, Stefanie Aeschbacher ∗,†, Sven Knecht ∗,†, Vincent Schlageter ∗, Philipp Krisai ∗,†, Patrick Badertscher ∗,†, Christian Sticherling ∗,†, Stefan Osswald ∗,†, Michael Kühne ∗,†,1,∗, Christine S Zuern ∗,†,1; Swiss-AF Investigators, on behalf of the
PMCID: PMC10123500  PMID: 37101946

Abstract

Background

Emerging evidence indicates that a high atrial fibrillation (AF) burden is associated with adverse outcome. However, AF burden is not routinely measured in clinical practice. An artificial intelligence (AI)-based tool could facilitate the assessment of AF burden.

Objective

We aimed to compare the assessment of AF burden performed manually by physicians with that measured by an AI-based tool.

Methods

We analyzed 7-day Holter electrocardiogram (ECG) recordings of AF patients included in the prospective, multicenter Swiss-AF Burden cohort study. AF burden was defined as percentage of time in AF, and was assessed manually by physicians and by an AI-based tool (Cardiomatics, Cracow, Poland). We evaluated the agreement between both techniques by means of Pearson correlation coefficient, linear regression model, and Bland-Altman plot.

Results

We assessed the AF burden in 100 Holter ECG recordings of 82 patients. We identified 53 Holter ECGs with 0% or 100% AF burden, where we found a 100% correlation. For the remaining 47 Holter ECGs with an AF burden between 0.01% and 81.53%, Pearson correlation coefficient was 0.998. The calibration intercept was -0.001 (95% CI -0.008; 0.006), and the calibration slope was 0.975 (95% CI 0.954; 0.995; multiple R2 0.995, residual standard error 0.017). Bland-Altman analysis resulted in a bias of -0.006 (95% limits of agreement -0.042 to 0.030).

Conclusion

The assessment of AF burden with an AI-based tool provided very similar results compared to manual assessment. An AI-based tool may therefore be an accurate and efficient option for the assessment of AF burden.

Keywords: Atrial fibrillation, Atrial fibrillation burden, Holter ECG, Artificial intelligence, Machine learning, Deep learning

Graphical abstract

graphic file with name fx1.jpg

Open in a new tab

AF = atrial fibrillation; AI = artificial intelligence; ECG = electrocardiogram.

Key Findings.

  • •

    The assessment of atrial fibrillation (AF) burden in 7-day Holter electrocardiogram (ECG) recordings with an artificial intelligence (AI)-based tool provided very similar results compared with the current gold standard of manual assessment by physicians.

  • •

    Physicians saved up to 2.5 hours per Holter ECG when Holter assessment was performed by the AI-based tool, especially if multiple AF episodes were present.

  • •

    Therefore, the use of an AI-based tool can be an accurate and time-saving alternative for AF burden assessment.

Introduction

Atrial fibrillation (AF) constitutes a major public health problem owing to the rising prevalence and its associated morbidity and mortality.1, 2, 3, 4 According to the guidelines of the European Society of Cardiology, AF is classified into 4 categories: paroxysmal, persistent, long-standing persistent, and permanent AF.3 However, this classification does not adequately reflect the time a patient actually spends in AF, especially in paroxysmal AF.5, 6, 7 The exact fraction of time spent in AF over a monitored period of time is termed AF burden.3,8 Emerging evidence indicates that a higher AF burden is associated with increased risk of adverse outcome, ie, new-onset heart failure, hospitalization for heart failure, stroke, and all-cause mortality.9, 10, 11 Nevertheless, the individual AF burden is not routinely measured in clinical practice and is therefore not systematically considered in the AF management.

Currently, the assessment of AF burden is not yet standardized. In clinical practice, AF burden is mostly measured manually by physicians from Holter electrocardiogram (ECG) recordings. This is time-consuming, especially in patients with multiple AF episodes. Therefore, assessment of AF burden based entirely on artificial intelligence (AI) without the need of manual over-read would be desirable and would facilitate the use of information on AF burden in clinical routine.

The aim of this analysis was to compare the AF burden measured by an AI-based tool trained previously in other cohorts with the current standard of manual assessment of AF burden in patients with known AF undergoing 7-day Holter ECG recordings.

Methods

We analyzed Holter ECG recordings from patients who were included in the Swiss-AF Burden study between March 2018 and May 2021. Swiss-AF Burden is a prospective, investigator-initiated multicenter study investigating the health consequences of AF burden. Swiss-AF Burden is embedded in the ongoing prospective, multicenter observational cohort study Swiss-AF.12 We performed 7-day 2-channel Holter ECG recordings at study entry and at the 1-year follow-up visit using a validated 3-channel 3-electrode monitoring device (Lifecard CF; Spacelabs Healthcare, Snoqualmie, WA) with a sampling rate of 1024 samples per second. Electrode 1 was attached to the upper border of the sternum, electrode 2 was fixed on the right mid-clavicular line at the lower right border of the ribcage, and electrode 3 was placed on the left anterior axillary line at the lower left border of the ribcage. Patients were allowed to take off the ECG recorder for short time periods only if absolutely necessary. Patients were instructed to perform all their daily activities as usual and to document symptoms and activities with the exact date and time on a log sheet.

Excluded from our analysis were patients with paced rhythms. We used all data available by May 2021. The study was approved by the local ethics committees, and all study participants gave their written informed consent.

Manual assessment of AF burden

ECG raw signals from anonymized Holter ECG recordings were uploaded into the Sentinel program and Pathfinder SL Software (Spacelabs Healthcare, Snoqualmie, WA) by USB port via a password-secured internet access. The data were backed up on a server of the University Hospital Basel. Furthermore, all participating centers created backups of their ECG recordings. All Holter ECG recordings were analyzed by 2 trained cardiologists and AF burden was assessed manually at a computer station. If the discrepancy in the manually detected AF burden values exceeded 5%, a third experienced cardiologist was consulted for a final decision. AF was defined according to current guidelines, ie, atrial fibrillation lasting more than 30 seconds.3 ECG Holter analysis was performed as in clinical routine using the following approach: First, the physicians screened for AF by visually analyzing heart rate plotted against time. With this strategy, AF can be recognized by its typical pattern, for example a sudden rise and drop in heart rate. Second, physicians went through the entire recording to detect AF episodes (e.g., by absence of p waves, phases of irregular heart rhythm). In a third step, tachycardias, bradycardias, supraventricular salvos, episodes with maximum and minimum heart rate, and the longest and shortest normal-to-normal and R-R intervals were reviewed manually. All AF episodes were labeled individually, and the duration of these AF episodes was summed up.

AF burden was calculated as the total duration of AF divided by the total duration of the recording time with an interpretable ECG signal (eg, no noise). Furthermore, we assessed whether AF episodes >24 hours were present. All symptoms recorded by the patient on the log sheet with date and time were checked in the ECG recording, and correlations with AF or any other arrhythmias were noted.

AI-based assessment of AF burden

For assessment of AF burden with an AI-based tool, we used the cloud-based Cardiomatics software (Cracow, Poland) for the detection of AF from Holter ECGs.13 The Cardiomatics tool is a CE-certified medical device (class IIa, certificate nos. 60148244 and 60148245), which provides automatic analysis of Holter recordings including AF detection. It is systematically verified in accordance with the ISO norm EN 60601-2-47:2015. Its algorithms were previously developed and trained on data derived from adult European cardiac patients (Holter recorders or ECG patches with 1–3 ECG channels from Germany, Switzerland, United Kingdom, and Poland).

The 100 Holter ECG recordings from 82 Swiss-AF Burden patients were exported in the ISHNE Holter standard output file format (.ecg) and uploaded on the Cardiomatics server, which was accessed via the Cardiomatics homepage (https://cardiomatics.com/). The Cardiomatics signal processing pipeline consists of the following steps: (1) signal preprocessing (filtering), (2) detection of QRS complexes, (3) classification of QRS complexes, (4) rhythm detection integrating information from both atrial and ventricular signals (including AF/atrial flutter), and (5) detection of nondiagnostic parts of the signal. The algorithm for AF detection is built based on a deep convolutional neural network. It takes the input of 12-second filtered segments of the signal from each channel and returns the probability of AF occurrence. The size of the input window allows the algorithm to generalize the characteristics of AF at different levels of detail. As a machine learning algorithm, it is trained on large sets of signals consisting of precise manual annotations (double checked and verified) of the AF episodes. AF occurrence is detected in the signal when the probability of AF occurrence predicted by the algorithm exceeds the decision threshold. The decision threshold is set during the precision-recall analysis on the internal testing sets containing a good representation of Holter recordings from different Holter devices. System operation, including the settings and quality control, is supervised by the Quality Assurance Team. The final results are presented in a web report available on the Cardiomatics platform (https://app.cardiomatics.com/).

Other study variables

We used standardized questionnaires to collect data on the patients’ medical history and medication. We classified AF in compliance with the AF guidelines of the European Society of Cardiology and then categorized AF into paroxysmal and nonparoxysmal AF.3 Body weight and height were obtained with calibrated devices and the body mass index was calculated.

Statistical analysis

Baseline characteristics were analyzed for the entire study population. Categorical variables are presented as numbers (percentages). Continuous variables are presented as mean ± standard deviation (SD) or median (interquartile range, IQR), as appropriate.

For Holter ECG recording with an AF burden >0% and <100%, we calculated the correlation between the manually assessed AF burden and the AI-assessed AF burden by means of the Pearson correlation coefficient. Taking the manual measurement as the gold standard, we provided a calibration plot for the AI-based tool. Because of 11 nonindependent Holter ECGs (baseline and follow-up Holter ECG of the same patient), we performed a sensitivity analysis with a linear mixed-effect model and included the patient ID as a random intercept. Agreement between the AF measurements was assessed by means of a Bland-Altman plot. We calculated the mean of the difference (bias) in AF burden and the 95% lower and upper limits of agreement (LoA). Furthermore, we assessed whether the correlation between manual and AI-based AF burden assessment differs across the spectrum of AF burden. Therefore, we divided the Holter recordings according to the median of AF burden and repeated the statistical analysis in these 2 groups.

All analyses were performed using R version 4.1.2 (2021-11-01, R Core Team).14

Results

A total of 100 ECGs from 82 patients participating in the Swiss-AF Burden study were available for this analysis. Table 1 shows the baseline characteristics of these patients. Mean age (± SD) was 70.2 ± 8.1 years, and 15 (18.3%) of them were female. Paroxysmal AF was present in 51 (62.2%) patients and nonparoxysmal AF in 31 (37.8%) patients. We identified 18 Holter ECGs with 0% and 35 Holter ECGs with 100% AF burden, where we found a 100% correlation between manual and AI-based AF burden assessment. AF burden was higher than 0% and lower than 100% in 47 Holter ECGs obtained in 36 patients (Figure 1), ranging between 0.01% and 81.53%. For the purpose of describing the performance of the AI-based tool to assess AF burden compared to manual measurement, the correlations from the 47 Holter ECG recordings with an AF burden >0% and <100% will be reported.

Table 1.

Baseline characteristics

Characteristics Overall study population
Number of patients N = 82
Age (years) 70.2 ± 8.1
Female sex 15 (18.3%)
Body mass index (kg/m2) 28.1 ± 5.2
Current smoker 7 (8.5%)
AF-related symptoms 58 (70.7%)
Heart rate (beats/min) 66 (58, 77)
Blood pressure (mmHg)
 Systolic 132 ± 18
 Diastolic 76 ± 10
AF type
 Paroxysmal 51 (62.2%)
 Nonparoxysmal 31 (37.8%)
CHA2DS2-VASc score 2.7 ± 1.7
History of PVI 17 (20.7%)
History of atrial flutter 21 (25.6%)
History of coronary artery disease 19 (23.2%)
History of stroke/TIA 14 (17.1%)
History of systemic embolism 4 (4.9%)
History of hypertension 48 (58.5%)
History of heart failure 13 (15.9%)
History of diabetes mellitus 8 (9.8%)
History of renal failure 10 (12.2%)
Antiarrhythmic agent
 Class IC 3 (3.7%)
 Class II 60 (73.2%)
 Class III 11 (13.4%)
Oral anticoagulation 71 (86.6%)
 Vitamin K antagonist 21 (25.6%)
 Direct oral anticoagulants 50 (61.0%)
Antiplatelet therapy 17 (20.7%)
Open in a new tab

Values are given as means ± standard deviation, median (interquartile range), or numbers (percentages). No missing values.

AF = atrial fibrillation; CHA2DS2-VASc = congestive heart failure, hypertension, age ≥75 years (2 points), diabetes, prior stroke or TIA or thromboembolism (2 points), vascular disease, age 65–74 years, female sex; PVI = pulmonary vein isolation; TIA = transient ischemic attack.

Figure 1.

Figure 1

Open in a new tab

Flowchart of study population and Holter electrocardiograms (ECGs). AF = atrial fibrillation. ∗Holter ECGs with an AF burden of 0% or 100% (where a 100% correlation between manual and artificial intelligence (AI)–measured AF burden was detected) were not used to assess the performance of the AI-based tool to detect AF burden.

Median duration of recordings was 168.4 hours (IQR 164.5; 168.5). An AF episode lasting longer than 24 hours was detected in 11 (23.4%) of the recordings. Median AF burden was 10.0% (IQR 2.7%; 32.4%) if assessed manually and 10.9% (IQR 2.4%; 32.8%) if assessed by the AI-based tool (Figure 2). Symptoms occurred simultaneously with an AF episode in 23 (48.9%) recordings. Symptoms included irregular pulse, sweating, malaise, reduced physical performance, palpitation, tachypnea, dyspnea, angina, vertigo, and fatigue. Symptoms correlated with other arrhythmias (supraventricular extrasystoles, ventricular extrasystoles, and supraventricular tachycardia) in 7 patients (14.9%).

Figure 2.

Figure 2

Open in a new tab

Violin and box plots of atrial fibrillation (AF) burden assessed manually and by artificial intelligence (AI)–based tool. Red cross indicates mean.

Figure 3 shows the association of AF burden assessed manually and AF burden determined by the AI-based tool. The Pearson correlation coefficient was 0.998. The linear regression model with the AI-based tool as a predictor of the manual measurement had a calibration intercept of -0.001 (95% CI -0.008; 0.006) and a calibration slope of 0.975 (95% CI 0.954; 0.995). Multiple R2 (coefficient of determination or goodness of fit) was 0.995, indicating that 99.5% of the variance in the set of manually assessed AF burden is explained by AF burden assessed by the AI-based tool. The residual standard error was 0.017. Fitting a linear mixed-effect model to account for the nonindependent ECGs of 11 patients showed no marked difference of the estimates (calibration intercept -0.001 [95% CI -0.008; 0.006] and calibration slope 0.975 [95% CI 0.952; 0.997]).

Figure 3.

Figure 3

Open in a new tab

Calibration plot of atrial fibrillation (AF) burden assessed manually and by artificial intelligence (AI)-based tool. Black line: 1 to 1 line; red line: linear regression line with calibration intercept = -0.001 (95% CI -0.008; 0.006) and calibration slope = 0.975 (95% CI 0.954; 0.995).

Bland-Altman analysis (Figure 4) resulted in a bias of -0.006 (95% LoA -0.042 [95% CI -0.051; -0.033] to 0.030 [95% CI 0.020; 0.039]). In 4 (8.5%) recordings, the difference between the 2 measurements exceeded the LoA.

Figure 4.

Figure 4

Open in a new tab

Bland-Altman plot of atrial fibrillation (AF) burden assessed manually and by artificial intelligence (AI)-based tool. Bias of -0.006 (red line) with 95% limits of agreement of -0.042 (95% CI -0.051; -0.033) to 0.030 (95% CI 0.020; 0.039) (dotted black lines). X-axis = mean of AF burden measurements, Y-axis = difference of the means of AF burden measurements assessed manually and by AI-based tool.

When dividing the 47 Holter recordings according to their median AF burden, we detected a slightly stronger correlation between manual and AI-based AF burden assessment in the recordings with AF burden higher than the median (Pearson correlation coefficient 0.995, calibration intercept -0.001 ([95% CI -0.021, 0.020], calibration slope = 0.975 [95% CI 0.931, 1.020]) vs lower than the median (Pearson correlation coefficient 0.974, calibration intercept 0.003 [95% CI -0.001, 0.008], calibration slope = 0.852 [95% CI 0.765, 0.938]).

The estimated time for 2 physicians to manually analyze 1 Holter ECG recording ranged between 1.5 hour and 2.5 hours in our study. The time for the export of the ECG raw signals and upload onto the AI server took about 5 minutes. The time required by the AI-based tool to generate a report for 7-day Holter ECGs was about 25–70 minutes. However, the delivery of the final report depends on the number of signals waiting for analysis, and for the AI-based tool we used equals approximately 2 hours for 24-hour Holter ECGs and 24 hours for Holter ECGs longer than 24 hours.

Discussion

In this study, we compared the AF burden obtained by manual state-of-the-art assessment by physicians with those obtained by an AI- and cloud-based tool for AF detection in 7-day Holter ECG recordings from AF patients. The main findings of our study are the following: First, we identified a 100% correlation of the AF burden assessed manually and by the AI-based tool in Holter ECGs with 0% and 100% AF burden. Second, for the Holter ECGs with an AF burden >0% and <100%, our results revealed a high correlation of the AI-based tool when compared with the AF burden obtained manually.

Currently, the AF burden is not measured routinely in AF patients owing to the time required to obtain measurements and owing to the lack of consequences according to current guidelines. However, the amount of AF burden has been shown to be associated with adverse outcomes in AF patients.11,15 In a study with 39,710 AF patients with an implanted cardiac device, increasing AF burden was strongly associated with a higher risk for new-onset heart failure, hospitalization for heart failure, and all-cause mortality.9 In the KP-RHYTHM study with 1,965 patients with paroxysmal AF, increasing AF burden, detected by a 14-day ECG patch, was strongly associated with a higher risk of ischemic stroke independent of known stroke risk factors.10 Therefore, the assessment of AF burden might allow for a more accurate risk assessment for adverse outcomes in AF patients.16 Furthermore, routine measurements of AF burden could help characterize the heterogeneous group of AF patients, including phenotype and disease progression.17 Finally, knowledge of the magnitude of AF burden could guide patient follow-up, therapy decision, and therapy success, but whether AF burden–guided patient management improves patient outcomes needs to be shown in future studies.18,19 So far, most data linking a higher AF burden to adverse outcome are derived from continuous monitoring data such as devices.9,15 However, in clinical routine, conventional Holter ECG recordings are mostly used for AF burden assessment, as they are easy to apply without any intervention for the patient.

Manual assessment of AF burden from Holter ECGs is time-consuming for physicians, especially if multiple AF episodes are present and if recordings are noisy. As the evaluation of AF burden is particularly difficult when the burden is >0% and <100%, we focused on such Holter ECGs to compare the AF burden derived from manual vs AI-based assessments. Overall, the AI tool matched human performance, as we found excellent agreement between AF burden from manual and AI-based assessment. The Pearson correlation coefficient of 0.998 reflects a high positive correlation. The calibration intercept was close to zero, and the calibration slope was close to 1 with a low residual standard error, indicating a strong association between the 2 types of assessments. The Bland-Altman analysis indicated that the AI-based tool minimally overestimated AF burden compared to the manual assessment (bias of -0.006), with a clinically acceptable LoA. When the ECG recordings were split by their median AF burden, the correlation between manual and AI-based AF burden assessment was slightly higher in the group of ECG recordings higher than the median. However, this difference might not be relevant in clinical practice.

Besides correlation, time efficiency should be considered as AI can speed up ECG analysis. In our analysis, physicians could save up to 2.5 hours per Holter ECG if the AI-based tool carried out the assessment. However, the time lag until the physician receives the report has to be taken into account, but this may be modifiable based on service level agreements. The measurement of AF burden is generally not urgent and may only influence long-term patient management. Therefore, this time lag is negligible. Another issue is data confidentiality, since the AI-based tool performs the analysis on a web-based cloud. With the current tool, this is addressed by complying with the General Data Protection Regulation ensuring the security of personal data processing.

AI is gaining a widespread interest in healthcare. However, in clinical routine, AI in ECG analysis is a relatively new technique that is still not used on a regular basis. This is in part owing to a “black box” concern and owing to possible sources of error. With regard to AI-based ECG analysis, sources of error could be that the definition of nondiagnostic parts of the ECG recordings may be different if analyzed by physicians vs analyzed by an AI tool. Especially, physicians might still be able to interpret the underlying rhythm (AF or not) if the signal is of poor quality, whereas AI algorithms may not work anymore. Second, the correct detection of supraventricular and ventricular extrasystoles is essential for AF detection, as sinus rhythm with multiple extrasystoles might be misinterpreted as AF.

The mechanism by which AF burden is associated with adverse outcome has not yet been entirely clarified. Despite the known association between AF and stroke, the exact pathophysiology remains unclear.3 Whether AF burden directly affects patient-specific outcomes (AF burden as a risk factor) or whether worse outcomes in patients with higher AF burden are attributable to more advanced comorbidities (AF burden as a risk marker), or both, is still controversial.3 In the ASSERT trial, only few patients had AF in the month before an embolic event.20 However, a recent study highlighted the temporal association between AF episodes and risk of stroke.21 Whether measuring AF burden can have an impact on clinical management of AF still needs to be determined. In cases where the AF guidelines are inconclusive (e.g., in patients with borderline CHA2DS2-VASc score), knowledge of AF burden might facilitate the decision-making process.22 The use of a tool based entirely on AI reduces physician workload and has the potential for the assessments of AF burden to be integrated in daily routine.

Study limitations

This study has several limitations. The main limitation is the small sample size, including a low number of women, which reduces the generalizability of our study. However, the duration of Holter ECG recordings was long, allowing for AF burden assessment over a period of 7 days. Our findings only apply to patients with confirmed AF and we cannot apply our findings to patients without diagnosed AF. Future studies should investigate the use of AI tools for AF detection in patients without known AF. The results of our study are only limited to the AI-based tool used, and each AI-based tool needs to be validated and calibrated individually. We included a small number of nonindependent Holter ECGs in our analysis. However, the sensitivity analysis did not reveal marked differences in the estimates when excluding these. Finally, various raters performed the first manual assessment of AF burden, which may increase the variability of our results, but the second rater was always the same physician, and a third rating was performed in case of discrepancy. Lastly, our findings only apply to the Swiss population, as the study was conducted in Switzerland.

Conclusion

The AI-based tool provided very similar results to manual analysis of AF burden by physicians in a real-world AF cohort. Thus, the use of an AI-based tool can be an accurate and efficient alternative for AF burden assessment and may have the potential to improve patient care.

Acknowledgments

We thank Rafał Samborski, Katarzyna Czosnyka, Katarzyna Barczewska, and Nikola Fajkis-Zajączkowska (Cardiomatics, Cracow, Poland) for their contribution (analysis of 100 Holter ECGs with the AI-based tool and description of methods of AI-based tool).

Funding Sources

The Swiss-AF Burden study is funded by the Swiss National Science Foundation (grant number 32473B_176178) and the Swiss Heart Foundation. The Swiss-AF study is funded by grants provided by the Swiss National Science Foundation (grant numbers 33CS30_148474, 33CS30_177520, 32473B_176178, and 32003B_197524), Swiss Heart Foundation, Foundation for Cardiovascular Research Basel, and the University of Basel. Holter ECGs were analyzed free of charge by Cardiomatics (Cracow, Poland). Cardiomatics had no role in the design and conduct of the study, data collection, statistical analysis, or data interpretation.

Disclosures

P. Badertscher received research funding from the University of Basel, the “Stiftung für Herzschrittmacher und Elektrophysiologie,” the “Freiwillige Akademische Gesellschaft Basel,” the Swiss Heart Foundation, and Johnson & Johnson, all outside the submitted work, and reports personal fees from Abbott, Boston Scientific, and Pfizer BMS. C. Sticherling is a member of the Advisory Board of Medtronic Europe and Advisory Board of Boston Scientific Europe and has received educational grants from Biosense Webster and Biotronik, research grants from the European Union’s FP7 program and Biosense Webster, and lecture and consulting fees from Abbott, Medtronic, Biosense Webster, Boston Scientific, Micro-Port, and Biotronik. S. Osswald received research grants from the Swiss National Science Foundation, Swiss Heart Foundation, Foundation for CardioVascular Research Basel, and Roche, and educational and speaker office grants from Roche, Bayer, Novartis, Sanofi, AstraZeneca, Daiichi-Sankyo, and Pfizer. S. Knecht received funding from the “Stiftung für Herzschrittmacher und Elektrophysiologie.” M. Kühne received personal fees from Bayer, Böhringer Ingelheim, Pfizer BMS, Daiichi Sankyo, Medtronic, Biotronik, Boston Scientific, Johnson & Johnson, and F. Hoffmann-La Roche Ltd, as well as grants from Bayer, Pfizer, Boston Scientific, BMS, Biotronik, and Daiichi Sankyo. C. S. Zuern received speaker fees from Vifor Pharma. The other authors have nothing to disclose.

Authorship

All authors attest they meet the current ICMJE criteria for authorship.

Patient Consent

All patients provided written informed consent.

Ethics Statement

The authors designed the study, gathered and analyzed the data according to the Helsinki Declaration guidelines on human research. The research protocol used in this study was reviewed and approved by the institutional review board.

Footnotes

ClinicalTrials.gov Identifier: NCT02105844

References

  • 1.Chugh S.S., Havmoeller R., Narayanan K., et al. Worldwide epidemiology of atrial fibrillation: a global burden of disease 2010 study. Circulation. 2014;129:837–847. doi: 10.1161/CIRCULATIONAHA.113.005119. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Colilla S., Crow A., Petkun W., Singer D.E., Simon T., Liu X. Estimates of current and future incidence and prevalence of atrial fibrillation in the U.S. adult population. Am J Cardiol. 2013;112:1142–1147. doi: 10.1016/j.amjcard.2013.05.063. [DOI] [PubMed] [Google Scholar]
  • 3.Hindricks G., Potpara T., Dagres N., et al. 2020 ESC Guidelines for the diagnosis and management of atrial fibrillation developed in collaboration with the European Association for Cardio-Thoracic Surgery (EACTS): The Task Force for the diagnosis and management of atrial fibrillation of the European Society of Cardiology (ESC) Developed with the special contribution of the European Heart Rhythm Association (EHRA) of the ESC. Eur Heart J. 2021;42:373–498. doi: 10.1093/eurheartj/ehaa612. [DOI] [PubMed] [Google Scholar]
  • 4.Kim M.H., Johnston S.S., Chu B.C., Dalal M.R., Schulman K.L. Estimation of total incremental health care costs in patients with atrial fibrillation in the United States. Circ Cardiovasc Qual Outcomes. 2011;4:313–320. doi: 10.1161/CIRCOUTCOMES.110.958165. [DOI] [PubMed] [Google Scholar]
  • 5.Boriani G., Pettorelli D. Atrial fibrillation burden and atrial fibrillation type: clinical significance and impact on the risk of stroke and decision making for long-term anticoagulation. Vascul Pharmacol. 2016;83:26–35. doi: 10.1016/j.vph.2016.03.006. [DOI] [PubMed] [Google Scholar]
  • 6.Charitos E.I., Pürerfellner H., Glotzer T.V., Ziegler P.D. Clinical classifications of atrial fibrillation poorly reflect its temporal persistence: insights from 1,195 patients continuously monitored with implantable devices. J Am Coll Cardiol. 2014;63:2840–2848. doi: 10.1016/j.jacc.2014.04.019. [DOI] [PubMed] [Google Scholar]
  • 7.Andrade J.G., Yao R.R.J., Deyell M.W., et al. Clinical assessment of AF pattern is poorly correlated with AF burden and post ablation outcomes: a CIRCA-DOSE sub-study. J Electrocardiol. 2020;60:159–164. doi: 10.1016/j.jelectrocard.2020.03.008. [DOI] [PubMed] [Google Scholar]
  • 8.Chen L.Y., Chung M.K., Allen L.A., et al. Atrial fibrillation burden: moving beyond atrial fibrillation as a binary entity: a scientific statement from the American Heart Association. Circulation. 2018;137:e623–e644. doi: 10.1161/CIR.0000000000000568. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Steinberg B.A., Li Z., O’Brien E.C., et al. Atrial fibrillation burden and heart failure: data from 39,710 individuals with cardiac implanted electronic devices. Heart Rhythm. 2021;18:709–716. doi: 10.1016/j.hrthm.2021.01.021. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Go A.S., Reynolds K., Yang J., et al. Association of burden of atrial fibrillation with risk of ischemic stroke in adults with paroxysmal atrial fibrillation: the KP-RHYTHM Study. JAMA Cardiol. 2018;3:601–608. doi: 10.1001/jamacardio.2018.1176. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Yang S.-Y., Huang M., Wang A.-L., et al. Atrial fibrillation burden and the risk of stroke: a systematic review and dose-response meta-analysis. World J Clin Cases. 2022;10:939–953. doi: 10.12998/wjcc.v10.i3.939. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Conen D., Rodondi N., Müller A., et al. Design of the Swiss Atrial Fibrillation Cohort Study (Swiss-AF): structural brain damage and cognitive decline among patients with atrial fibrillation. Swiss Med Wkly. 2017;147 doi: 10.4414/smw.2017.14467. [DOI] [PubMed] [Google Scholar]
  • 13.Krzowski B., Rokicki J., Główczyńska R., et al. The use of machine learning algorithms in the evaluation of the effectiveness of resynchronization therapy. J Cardiovasc Dev Dis. 2022;9:17. doi: 10.3390/jcdd9010017. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.R Core Team . R Foundation for Statistical Computing; Vienna, Austria: 2021. R: A language and environment for statistical computing.http://www.r-project.org/ [Internet] [Google Scholar]
  • 15.Brachmann J., Sohns C., Andresen D., et al. Atrial fibrillation burden and clinical outcomes in heart failure: the CASTLE-AF Trial. JACC Clin Electrophysiol. 2021;7:594–603. doi: 10.1016/j.jacep.2020.11.021. [DOI] [PubMed] [Google Scholar]
  • 16.Tiver K.D., Quah J., Lahiri A., Ganesan A.N., McGavigan A.D. Atrial fibrillation burden: an update-the need for a CHA2DS2-VASc-AFBurden score. Europace. 2021;23:665–673. doi: 10.1093/europace/euaa287. [DOI] [PubMed] [Google Scholar]
  • 17.Watanabe E., Inoue H., Atarashi H., et al. Clinical phenotypes of patients with non-valvular atrial fibrillation as defined by a cluster analysis: a report from the J-RHYTHM registry. Int J Cardiol Heart Vasc. 2021;37 doi: 10.1016/j.ijcha.2021.100885. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Andrade J.G., Wells G.A., Deyell M.W., et al. Cryoablation or drug therapy for initial treatment of atrial fibrillation. N Engl J Med. 2021;384:305–315. doi: 10.1056/NEJMoa2029980. [DOI] [PubMed] [Google Scholar]
  • 19.Waks J.W., Passman R.S., Matos J., et al. Intermittent anticoagulation guided by continuous atrial fibrillation burden monitoring using dual-chamber pacemakers and implantable cardioverter-defibrillators: results from the Tailored Anticoagulation for Non-Continuous Atrial Fibrillation (TACTIC-AF) pilot study. Heart Rhythm. 2018;15:1601–1607. doi: 10.1016/j.hrthm.2018.06.027. [DOI] [PubMed] [Google Scholar]
  • 20.Brambatti M., Connolly S.J., Gold M.R., et al. Temporal relationship between subclinical atrial fibrillation and embolic events. Circulation. 2014;129:2094–2099. doi: 10.1161/CIRCULATIONAHA.113.007825. [DOI] [PubMed] [Google Scholar]
  • 21.Singer D.E., Ziegler P.D., Koehler J.L., Sarkar S., Passman R.S. Temporal association between episodes of atrial fibrillation and risk of ischemic stroke. JAMA Cardiol. 2021;6:1364–1369. doi: 10.1001/jamacardio.2021.3702. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Kaplan R.M., Koehler J., Ziegler P.D., Sarkar S., Zweibel S., Passman R.S. Stroke risk as a function of atrial fibrillation duration and CHA(2)DS(2)-VASc score. Circulation. 2019;140:1639–1646. doi: 10.1161/CIRCULATIONAHA.119.041303. [DOI] [PubMed] [Google Scholar]

Articles from Cardiovascular Digital Health Journal are provided here courtesy of Elsevier

RESOURCES